On Sat, Oct 20, 2018 at 1:01 AM Nicklas Karlsson < nicklas.karlsso...@gmail.com> wrote:
> (1.) > Then using state space model and integrator I want to integrate error > between where signal is expected to be and there measurement suggest it is. > As is now I integrate error between reference signal measured output and > then gain is reduced to get a more stable system the integrator will add an > overshoot. Basically my idea is to integrate only model error. > > (2.) > For a linear state space model there must be a "standardized" method to > calculate feed forward. This is of particular interest then gain is reduced > to get a more stable system. > > (1)Integrators almost always introduce an overshoot. I am pretty sure you always want to integrate the difference between where you are and where you want to be (reference). Adding an integrator on the difference between model error and measurement sounds a lot like "residual whitening." A Kalman filter can be set up to estimate unknown model parameters. The unknown parameters are set up as states of the system model. (2)I think most people set FF gains from system knowledge and previous experience. There are optimal control techniques that will calculate them for you. H infinity and related control schemes. The Julia language has toolboxes for Kalman filters (and variations) and control theory. My guess is that you can find a system model that matches your machine if you look. Might require a friend at a university to d/l some papers for you. Eric Keller Boalsburg, Pennsylvania _______________________________________________ Emc-users mailing list Emc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-users
